Previous work on multimodal machine translation has shown that visual information is only needed in very specific cases, for example in the presence of ambiguous words where the textual context is not sufficient. As a consequence, models tend to learn to ignore this information. We propose a translate-and-refine approach to this problem where images are only used by a second stage decoder. This approach is trained jointly to generate a good first draft translation and to improve over this draft by (i) making better use of the target language textual context (both left and right-side contexts) and (ii) making use of visual context. This approach leads to the state of the art results. Additionally, we show that it has the ability to recover from erroneous or missing words in the source language.

Mental health problems represent a major public health challenge. Automated analysis of text related to mental health is aimed to help medical decision-making, public health policies and to improve health care. Such analysis may involve text classification. Traditionally, automated classification has been performed mainly using machine learning methods involving costly feature engineering. Recently, the performance of those methods has been dramatically improved by neural methods. However, mainly Convolutional neural networks (CNNs) have been explored. In this paper, we apply a hierarchical Recurrent neural network (RNN) architecture with an attention mechanism on social media data related to mental health. We show that this architecture improves overall classification results as compared to previously reported results on the same data. Benefitting from the attention mechanism, it can also efficiently select text elements crucial for classification decisions, which can also be used for in-depth analysis.

Contextualised word embeddings is a powerful tool to detect contextual synonyms. However, most of the current state-of-the-art (SOTA) deep learning concept extraction methods remain supervised and underexploit the potential of the context. In this paper, we propose a self-supervised pre-training approach which is able to detect contextual synonyms of concepts being training on the data created by shallow matching. We apply our methodology in the sparse multi-class setting (over 15,000 concepts) to extract phenotype information from electronic health records. We further investigate data augmentation techniques to address the problem of the class sparsity. Our approach achieves a new SOTA for the unsupervised phenotype concept annotation on clinical text on F1 and Recall outperforming the previous SOTA with a gain of up to 4.5 and 4.0 absolute points, respectively. After fine-tuning with as little as 20% of the labelled data, we also outperform BioBERT and ClinicalBERT. The extrinsic evaluation on three ICU benchmarks also shows the benefit of using the phenotypes annotated by our model as features.

A serious obstacle to the development of Natural Language Processing (NLP) methods in the clinical domain is the accessibility of textual data. The mental health domain is particularly challenging, partly because clinical documentation relies heavily on free text that is difficult to de-identify completely. This problem could be tackled by using artificial medical data. In this work, we present an approach to generate artificial clinical documents. We apply this approach to discharge summaries from a large mental healthcare provider and discharge summaries from an intensive care unit. We perform an extensive intrinsic evaluation where we (1) apply several measures of text preservation; (2) measure how much the model memorises training data; and (3) estimate clinical validity of the generated text based on a human evaluation task. Furthermore, we perform an extrinsic evaluation by studying the impact of using artificial text in a downstream NLP text classification task. We found that using this artificial data as training data can lead to classification results that are comparable to the original results. Additionally, using only a small amount of information from the original data to condition the generation of the artificial data is successful, which holds promise for reducing the risk of these artificial data retaining rare information from the original data. This is an important finding for our long-term goal of being able to generate artificial clinical data that can be released to the wider research community and accelerate advances in developing computational methods that use healthcare data.

Simultaneous machine translation (SiMT) aims to translate a continuous input text stream into another language with the lowest latency and highest quality possible. The translation thus has to start with an incomplete source text, which is read progressively, creating the need for anticipation. In this paper, we seek to understand whether the addition of visual information can compensate for the missing source context. To this end, we analyse the impact of different multimodal approaches and visual features on state-of-the-art SiMT frameworks. Our results show that visual context is helpful and that visually-grounded models based on explicit object region information are much better than commonly used global features, reaching up to 3 BLEU points improvement under low latency scenarios. Our qualitative analysis illustrates cases where only the multimodal systems are able to translate correctly from English into gender-marked languages, as well as deal with differences in word order, such as adjective-noun placement between English and French.